PolyEffNetV1: A CNN based colorectal polyp detection in colonoscopy images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal polyps are a primary cause of colorectal cancer.
• Early polyp detection via colonoscopy is crucial for effective treatment.
• Variations in polyp size and shape present challenges in automated detection.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for polyp segmentation and classification in colonoscopy images.
• To improve the accuracy and efficiency of polyp identification in endoscopic procedures.

Main Methods:

• Proposed PolypEffNetV1, a hybrid model combining U-Net for segmentation and EfficientNetB5 for classification.
• Utilized KVASIR dataset (1000 images) for segmentation and a combined KVASIR/CVC dataset (1612 images) for classification.
• Trained models to differentiate between polyps and non-polyp inflamed regions.

Main Results:

• PolypEffNetV1 achieved 97.1% testing accuracy, 0.84 Jaccard index, 0.91 dice coefficient, and 0.89 F1-score for segmentation.
• The classification model demonstrated 99% validation accuracy, 98% specificity, and 99% sensitivity.
• High performance metrics indicate robust polyp detection capabilities.

Conclusions:

• The developed deep learning system effectively segments and classifies polyps in colonoscopy images.
• PolypEffNetV1 offers a valuable tool for gastroenterologists, enhancing diagnostic accuracy and patient outcomes.
• The models are suitable for real-time edge deployment or integration into existing clinical software for improved healthcare quality.